|Home | About | Journals | Submit | Contact Us | Français|
Eukaryotic initiation factor 4G (eIF4G) promotes mRNA recruitment to the ribosome by binding to the mRNA cap- and poly(A) tail-binding proteins eIF4E and Pap1p. eIF4G also binds eIF4A at a distinct HEAT domain composed of five stacks of antiparallel α-helices. The role of eIF4G in the later steps of initiation, such as scanning and AUG recognition, has not been defined. Here we show that the entire HEAT domain and flanking residues of Saccharomyces cerevisiae eIF4G2 are required for the optimal interaction with the AUG recognition factors eIF5 and eIF1. eIF1 binds simultaneously to eIF4G and eIF3c in vitro, as shown previously for the C-terminal domain of eIF5. In vivo, cooverexpression of eIF1 or eIF5 reverses the genetic suppression of an eIF4G HEAT domain Ts− mutation by eIF4A overexpression. In addition, excess eIF1 inhibits growth of a second eIF4G mutant defective in eIF4E binding, which was also reversed by cooverexpression of eIF4A. Interestingly, excess eIF1 carrying the sui1-1 mutation, known to relax the accuracy of start site selection, did not inhibit the growth of the eIF4G mutant, and sui1-1 reduced the interaction between eIF4G and eIF1 in vitro. Moreover, a HEAT domain mutation altering eIF4G moderately enhances translation from a non-AUG codon. These results strongly suggest that the binding of the eIF4G HEAT domain to eIF1 and eIF5 is important for maintaining the integrity of the scanning ribosomal preinitiation complex.
Ribosomal recognition of the first AUG codon in the mRNA is a distinctive feature of canonical translation initiation in eukaryotes. During eukaryotic translation initiation, the small (40S) ribosomal subunit binds the eukaryotic initiation factor 2 (eIF2)/GTP/methionyl initiator tRNA (Met-tRNAiMet) ternary complex to form the 43S preinitiation complex. Subsequent joining of mRNA, with eIF4F bound to its m7G cap and poly(A)-binding protein (PABP) bound to its poly(A) tail, produces the 48S preinitiation complex. The eIF3 stimulates recruitment of Met-tRNAiMet and mRNA to the 40S ribosome (for review, see reference 14). Base pairing between the anticodon of Met-tRNAiMet and triplet sequences in the mRNA is monitored with the help of the small factors eIF1 and eIF1A, as the 48S complex migrates from the 5′ end of the mRNA during the process called scanning. Correct base pairing between the Met-tRNAiMet anticodon and the first AUG codon triggers the hydrolysis of GTP bound to eIF2 in a reaction involving the GTPase-activating protein eIF5, which in turn leads to dissociation of preassembled eIFs and formation of an initiation complex containing the AUG-anticodon base pair in the ribosomal P-site. eIF5B then facilitates the joining of the 40S initiation complex with the 60S subunit to form the 80S initiation complex, the direct precursor for the elongation of polypeptide chains on the ribosome.
The eIF4G subunit of eIF4F plays a pivotal role in mRNA recruitment to the ribosome, as it concurrently binds mRNA and the 40S ribosome through linker proteins (13, 34). Studies on both mammalian and yeast eIF4G indicate that the N-terminal domain of eIF4G contains the conserved binding sites for the cap-binding subunit of eIF4F, eIF4E, and PABP (15, 17, 31, 32, 35), whereas the middle domain contains the conserved binding site for the helicase eIF4A (19, 24, 26). The structure of the middle domain of mammalian eIF4GII was solved at atomic resolution and found to contain a type of HEAT domain with five pairs of antiparallel α-helices stacked together to form a crescent-shaped platform (21). In addition to the N-terminal and middle domains, mammalian eIF4G contains a C-terminal domain with a second eIF4A binding site (24) and the binding site for the eIF4E kinase Mnk (33). Yeast eIF4G lacks this C-terminal domain.
It remains unclear how eIF4G is linked to the 40S ribosome and whether or not it plays a role in steps following mRNA binding of the 48S complex, especially scanning and AUG recognition. It has not even been determined whether eIF4G leaves the ribosomal initiation complex, once the latter binds the 5′ cap of the message, or stays on the ribosome to stimulate the scanning process. In the case of mammals, eIF3 can bind directly to the eIF4G middle domain (18, 24), such that eIF3 can be considered a linker protein for mRNA recruitment. For Saccharomyces cerevisiae, however, eIF3 has not yet been reported to bind directly to eIF4G. Instead, the C-terminal domain (CTD) of eIF5 can bind simultaneously to eIF3 and eIF4G, thereby bridging the eIF3-eIF4G interaction, at least in vitro (6). Hence, the linkage of eIF4G to the rest of the 48S complex may involve multiple interactions, and it is possible that eIF4G plays a role in scanning, AUG recognition, or 60S subunit joining as an integral part of the 48S complex.
Genetic and biochemical analyses of yeast Sui− (suppressor of initiation codon mutation) mutations suggest that a higher rate of GTP hydrolysis on eIF2 reduces the accuracy of AUG selection in the Sui− mutants (16). As Sui− mutations have been mapped to eIF1, the N-terminal domain of eIF5, and all three subunits of eIF2, this group of eIFs may be juxtaposed in the initiation complex to form a recognition site for codon-anticodon base pairing that couples AUG selection to GTP hydrolysis. Our previous studies indicated that the c-subunit of eIF3 (Nip1p) and the eIF5 CTD mediate formation of the multifactor complex (MFC) containing eIFs 1, 2, 3, and 5 and Met-tRNAiMet, thereby promoting the assembly of these factors during the scanning and AUG recognition processes (3, 6).
In this paper, we show that yeast eIF1 and eIF5 interact with the middle HEAT domain (and flanking residues) of eIF4G in vitro and in vivo. By characterizing temperature-sensitive mutants of yeast eIF4G, we also obtained evidence suggesting that eIF4G is involved in the AUG recognition process. Moreover, a known Sui− mutation in eIF1 reduced its binding to eIF4G both in vivo and in vitro. Our results favor the model that eIF4G stays on the 40S ribosome until the ribosome encounters the first AUG in the capped mRNA.
Plasmids and yeast strains used in this study are listed in Tables Tables11 and and2,2, respectively. Details of their construction are available upon request. eIF4G1 (His6 tagged) was expressed in SF9 insect cells and purified as described previously (38). eIF4G21-513 and eIF4F2439-514 (both His6 tagged) were expressed in Escherichia coli RIL (Stratagene) transformants carrying pHis-4G2ΔN and pHis-4G2ΔS, respectively, and purified by Ni2+ affinity resin followed by chromatography on a Hitrap heparin column (Amersham Pharmacia), using a 0-to-1 M KCl gradient for elution. Glutathione S-transferase (GST)-eIF1 and GST-eIF5 were expressed in E. coli, purified by glutathione-Sepharose resin (Amersham Pharmacia), and eluted with glutathione, which was removed by dialysis. Expression and purification of eIF3c1-156 (His6 tagged; known as His-NIP1-N) were described elsewhere (6). Expression and purification of eIF1 and eIF4A (both His6 tagged) by Ni2+ affinity and ion-exchange column purification are described elsewhere (39).
Microtiter plate binding assays were performed on EIA/RIA plates (96-well polystyrene plates; Corning Inc., Corning, N.Y.). Protein samples were immobilized to the well by incubating the solution in 100 μl of immobilization buffer (50 mM NaHCO3 [pH 9.6], 5 mM β-mercaptoethanol [BME]) at 4°C for 4 to 6 h. The immobilization buffer was removed, and the well surface was blocked with 200 μl of 1% gelatin in HBS (20 mM HEPES [pH 7.5], 100 mM KCl, 5 mM MgCl2, 5 mM BME) per well for 1 h at 37°C. For the binding reaction, the binding partner was mixed with primary and secondary antibodies at appropriate dilutions and applied to the respective wells in 100 μl of HBS containing 0.2% gelatin. Plates were then incubated at 4°C overnight. Each well was then washed three times with 200 μl of wash buffer (0.05% Tween 20 in HBS) and stained with 100 μl of staining solution (1 mg of 4-nitrophenyl phosphate [Sigma]/ml in 50 mM NaHCO3 [pH 9.6], 5 mM MgCl2). Plates were left to stain until the strongest wells showed a clear yellow color, and the resulting optical density at 405 nm (OD405) was determined in a microplate reader (Bio-Rad). Antibody dilutions used were as follows: anti-GST, 1:3,000; anti-eIF4G1, 1:1,000; anti-rabbit alkaline phosphatase conjugate (Sigma), 1:10,000. As shown below in Fig. Fig.1A1A and B, primary antibodies did not inhibit the binding reaction because the employed antibodies were raised against the protein segment outside of the binding domain (anti-eIF4G1 against the N-half of eIF4G1; anti-GST against the GST portion of the fusion proteins).
Coimmunoprecipitation with antihemagglutinin (anti-HA) antibodies was conducted essentially as described previously (5, 35). Briefly, whole-cell extracts (WCE) were prepared from yeast transformants carrying appropriate URA3 plasmids, grown in SC-ura medium to an OD600 of ~1 by glass bead disruption in 0.5× 4GA buffer (50 mM potassium acetate, 1 mM magnesium acetate, 0.5 μg of leupeptin/ml, 0.5 μg of pepstatin/ml, 0.5 mM phenylmethylsulfonyl fluoride, 7 mM BME, and 15 mM HEPES [pH 7.5]). Two-hundred micrograms of extract was adjusted to 0.1% Triton X-100 and 0.01% sodium dodecyl sulfate (SDS) and incubated for 1 h at 4°C with monoclonal anti-HA antibodies (BAbCO) attached to protein A-Sepharose (Amersham Pharmacia Biotech). Immune complexes were analyzed by Western blotting as described previously (5).
GST pull-down assays with 35S-labeled proteins, synthesized in a rabbit reticulocyte lysate, were conducted as described previously (5). The amounts of bound 35S-labeled proteins were quantitated with STORM or TYPHOON (Molecular Dynamics). Dissociation constants (Kd) were deduced from the fraction of 35S-labeled protein (f) bound to a known molar concentration of GST fusion protein (G) from the following equation: Kd = G[(1/f) − 1]. This holds only when G is >>[35S-protein].
Dissociation constants for the interaction between eIF4G2453-914 and eIF4A were determined more accurately with eIF4G2453-914 covalently immobilized on B1 sensor chips, using the BIAcore system. eIF4A was allowed to interact with the immobilized protein in HBS at a flow rate of 50 μl/min. Equilibrium response levels were plotted against eIF4A concentrations, and the Kd was determined using BIAevaluation software (version 3.0.2).
We previously found that both yeast eIF4G isoforms eIF4G1 and eIF4G2 could bind eIF5 in vitro and localized the eIF5 binding domain of eIF4G2 to its C-terminal half, which is homologous to the middle domain of mammalian eIF4G containing HEAT motifs (6). To examine if yeast eIF4G binds eIF1, we first conducted in vitro protein binding assays. Purified yeast eIF1 or bovine serum albumin (BSA) as a control was immobilized on a microtiter plate and incubated with recombinant yeast eIF4G1 purified from insect cells. Binding of eIF4G1 to the immobilized proteins was detected via anti-eIF4G1 antibodies and alkaline phosphatase-coupled secondary antibodies as described in Materials and Methods. As shown in Fig. Fig.1A,1A, eIF4G1 specifically bound to the immobilized eIF1, but not BSA, on the microtiter plate. To examine which domain of eIF4G binds eIF1, N- and C-terminal halves of eIF4G2 (eIF4G21-513 or eIF4G2439-914) containing the eIF4E/PABP- or eIF4A-binding domain, respectively, were purified and adsorbed to the plate. Binding of GST-fused derivatives of eIF1 or eIF5 was then detected via anti-GST antibodies. As shown in Fig. Fig.1B,1B, GST-eIF1 bound specifically to eIF4G2439-914 (column 6), but not to eIF4G21-513 (column 5). Similar results were obtained for GST-eIF5 (columns 8 and 9), as observed previously (6).
We also tested whether the binary complexes observed in Fig. Fig.1B1B were stable enough to be isolated by affinity purification. eIF4G2439-914 was mixed with purified GST alone, GST-eIF1, or GST-eIF5, and the resulting complexes were repurified on a glutathione-Sepharose column and visualized by silver staining. As shown in Fig. Fig.1C,1C, bottom panel, the eIF4G2 segment specifically copurified with GST-eIF1 (lane 2) and GST-eIF5 (lane 4) but not with GST alone (lane 1). Treatment of the binary complexes with RNase A did not abolish these interactions (lanes 3 and 5), excluding the possibility that they are mediated by RNA bound fortuitously to a component of the binding reaction. We conclude that the C-terminal half of eIF4G, encompassing the HEAT domain, can bind directly to both eIF1 and eIF5.
The ratios of eluted GST fusion protein to coeluted eIF4G439-915 suggest that the interactions involved in tethering the eIF4G fragment to GST-eIF1 or GST-eIF5 are relatively weak. Since all proteins for the experiment presented in Fig. Fig.1C1C were used at initial concentrations of 1.25 μM, the results suggest an apparent equilibrium dissociation constant (Kd) of ~10 μM. In further experiments, we also used different amounts of GST-eIF1 and GST-eIF5241-405 (the minimal segment for eIF4G binding, known as B6 ) tethered to the glutathione resin for the binding assays with 35S-labeled eIF4G439-915 (35S-eIF4G439-915) (Fig. (Fig.1D,1D, top panel). Plotting the fraction of bound 35S-eIF4G439-915 against the concentration of GST fusion protein was consistent with a hyperbolic curve (Fig. (Fig.1D),1D), and apparent Kd values were deduced to be 9.4 ± 2.3 μM for GST-eIF1 and 8.2 ± 1.8 μM for GST-eIF5241-405. It should be noted, however, that these calculations neglect the loss of the eIF4G2 fragment that occurs during the washing steps. These values therefore represent upper limits for the Kd value of the respective interactions.
The top panel of Fig. Fig.2A2A shows the location of the 10 α-helices that constitute the HEAT domain of yeast eIF4G2 between residues 557 and 812. Two consecutive α-helices (e.g., 1a and 1b) correspond to a single HEAT repeat and form one pair of antiparallel helices as a part of the entire structure (2). The eIF4G2439-914 segment used in the previous experiments contains additional segments flanking the deduced HEAT domain (Fig. (Fig.2A,2A, row 2). To narrow down the binding domains for eIF1 and eIF5, we tested GST-eIF1 and GST-eIF5241-405 for binding to differently truncated eIF4G2 segments synthesized and labeled with [35S]methionine in a rabbit reticulocyte lysate. Figure Figure2A2A shows the results of these experiments.
First, we found that 35S-eIF4G2439-914 (row 2), but not 35S-eIF4G21-513 (row 1), efficiently bound to both GST-eIF1 and GST-eIF5241-405, confirming that the C-terminal half of eIF4G is responsible for binding to eIF1 or eIF5. Further N-terminal truncation to residue 514 or beyond abolished the binding to both proteins (rows 3 and 4). This result placed the N-terminal boundary of the binding domains for eIF1 and eIF5 between residues 440 and 513, N terminal to all of the HEAT repeats (see Discussion). The C-terminal deletions in eIF4G2439-914 extending up to residue 577, between the first and second HEAT repeats (rows 5 to 7), reduced binding of the eIF4G2 peptides to GST-eIF5241-405 by 40% or less. However, the same C-terminal deletions (rows 5 to 7) reduced the binding of eIF4G2 to GST-eIF1 by 70 to 88% of that given by eIF4G2439-914 (row 2). Thus, the segment C terminal to the HEAT repeats is required for optimal binding of both eIF5 and eIF1 but is particularly important for eIF1 binding. It is interesting that the smallest eIF4G2439-577 segment still showed a weakened but specific binding to GST-eIF1 and GST-eIF5241-405 (row 7), suggesting that the primary eIF1/eIF5-binding site may be located within the N-terminal third of eIF4G2439-914.
The tif4632-1, tif4632-6, and tif4632-8 mutations alter four, two, and four amino acids, respectively, in the C-terminal half of eIF4G2 (Fig. (Fig.2B)2B) and have been shown to reduce its binding to eIF4A in vitro and in vivo (26). According to the structure solved for the mammalian eIF4GII HEAT domain, most, if not all, of these mutations map in the hydrophobic core and almost certainly destabilize protein structure (21). Thus, we used these mutations to examine the effect of disrupting the entire HEAT domain structure. We found that all these mutations reduced the binding of eIF4G2439-914 to eIF1 and eIF5 (rows 8 to 10). In particular, tif4632-1, altering the entire HEAT domain, reduced the interaction with eIF5 by 81% (row 8) and tif4632-8, altering the N-terminal extension and a part of the HEAT domain, reduced both the interactions with eIF1 and eIF5 by 75 to 78% (row 10). Based on all these results, we conclude that the entire HEAT repeat domain and flanking residues between 439 and 914 are required for optimal binding of both eIF1 and eIF5.
We reported previously that eIF1 and eIF5 can interact simultaneously with eIF3c (3). Having observed that eIF1 and eIF5 interact with the eIF4G domain at binding sites close to each other (Fig. (Fig.2),2), we wished to examine whether eIF1 and eIF5 can bind simultaneously to eIF4G. We also asked whether eIF1 could bind simultaneously to eIF4G and eIF3c, as shown previously for the eIF5 CTD (6).
To test if eIF1 and eIF5 can bind to eIF4G2 simultaneously, we first examined if the purified eIF4G2439-914 can serve as a bridge between eIF1 and eIF5 (Fig. (Fig.3A).3A). We used a purified eIF3c segment (eIF3c1-156), known to form a bridge between eIF1 and eIF5, as a positive control in the present experiments. As shown in Fig. Fig.3B,3B, lanes 5 and 6, we found that eIF4G2439-914 did not enhance the relatively weak interaction between GST-eIF5241-405 and 35S-eIF1 (bottom panel), even though the purified eIF4G segment can bind efficiently to GST-eIF5241-405 (top panel). By contrast, purified eIF3c1-156 enhanced the GST-eIF5241-405/eIF1 interaction as reported previously (Fig. (Fig.3B,3B, lanes 9 and 10). The weak interaction between eIF5 and eIF1 was not prevented by the binding of eIF4G to eIF5 (lanes 5 and 6), suggesting that the eIF5 CTD can bind concurrently to eIF1 and eIF4G.
We then examined a reverse combination regarding which factor is fused to GST and present in excess. As shown in lane 5 of Fig. Fig.3C,3C, the N-terminal fusion of eIF1 to a GST-fused protein abolished its ability to bind eIF5 (manuscript in preparation for our detailed analyses on the eIF1/eIF5 interaction [C. R. Singh, H. He, M. Ii, and K. Asano, unpublished data]). We found that the eIF3c segment, but not the eIF4G2 segment, bridged the interaction between GST-eIF1 and 35S-eIF5241-405 (Fig. (Fig.3C,3C, lanes 6 and 10), indicating again that the eIF4G segment does not serve as a bridge between eIF1 and eIF5.
The amount of the eIF4G2 segment (~2 μg) tethered to GST-eIF5241-405 or -eIF1 (lanes 6 in top panels of Fig. Fig.3B3B and C) should be sufficient to bind 35S-eIF1 or -eIF5241-405, respectively, at a detectable level if the eIF4G2 segment can bind to the 35S-labeled proteins at an affinity similar to those deduced in Fig. Fig.1D.1D. Therefore, the results in Fig. Fig.3B3B and C even suggest that eIF1 and eIF5 cannot bind to eIF4G2 simultaneously. To test this idea directly, we conducted the binding assay between GST-eIF4G2439-914 and 35S-eIF5241-405 in the presence of different amounts of eIF1. As shown in Fig. Fig.3D,3D, the presence of eIF1 in ~10-fold molar excess over GST-eIF4G2439-914 inhibited this interaction (lane 4). Thus, the interactions of the eIF4G2 HEAT domain with eIF1 and eIF5 appear to be mutually exclusive.
To test if eIF4G and eIF3c can bind to eIF1 simultaneously, we allowed GST-eIF3c1-156 to bind 35S-eIF4G2439-914 in the presence of recombinant eIF1. As shown in Fig. Fig.3E,3E, eIF1 efficiently bound to GST-eIF3c1-156 (top panel) and formed a bridge between eIF3c and eIF4G2 (bottom panel). We also found that 35S-eIF4G2439-914 bound efficiently to GST-eIF1 in the presence of eIF3c1-156 in an amount 10-fold over that of GST-eIF1 (data not shown). Therefore, eIF4G and eIF3c can bind to eIF1 simultaneously. The results shown in Fig. Fig.33 suggest that eIF1 and eIF5 are tethered to eIF4G at different steps in the initiation pathway, while they are held together by the simultaneous interactions with eIF3c (see Discussion).
Next, we examined if eIF1 interacts with eIF4G in vivo. We previously reported that eIF1, eIF5, eIF2, and a small proportion of eIF4G1 immunoprecipitated with the HA epitope-tagged eIF3i (Tif34p) subunit (3, 6). However, the observed eIF4G interactions may be mediated by the 40S ribosomes to which HA-eIF3 likely binds. The small amount of eIF4G immunoprecipitated with HA-eIF3 suggests that the steady-state level of 48S complexes is very low compared to 43S complexes containing only the MFC components (3). In order to increase the fraction of eIF1/eIF4G complexes free of the 40S ribosome, we overproduced eIF1 in a strain encoding HA-tagged eIF4G1 and tested if the overproduced eIF1 coimmunoprecipitated with anti-HA antibodies. As shown in Fig. Fig.4A,4A, top panel, HA-eIF4G1 was specifically precipitated regardless of overexpression of eIF1 (top panel, lanes 2, 5, 8, and 11) together with the eIF4E subunit of eIF4F (Fig. (Fig.4A,4A, second panel). Importantly, eIF1, when overproduced, was specifically precipitated with HA-eIF4G1 (compare lanes 8 and 11). The amount of coimmunoprecipitated eIF1 corresponds to ~10% of the native level of eIF1 (compare lanes 7 and 11). As a negative control, immunoblotting with anti-eIF3g antibodies suggested that little or no eIF3 is associated with HA-eIF4G1 under these conditions (Fig. (Fig.4A,4A, bottom panel).
Using a yeast strain encoding HA-eIF4G2 as a sole source of eIF4G, we also found that a small but significant amount of eIF1 (~5% of the native level) precipitated with anti-HA antibodies, only when eIF1 was overexpressed (data not shown). Therefore, both eIF4G1 and eIF4G2 can specifically bind to eIF1 when the eIF1 concentration is raised in the cell. Because we did not detect the association of eIF1 with HA-eIF4G at the native level of eIF1, the observed interactions may not be physiological. However, we show below that overexpression of eIF1, and hence its increased interaction with eIF4G, has a physiological impact on the growth of some eIF4G mutants under certain conditions (see Fig. Fig.5,5, below). Likewise, we found that eIF5 coprecipitated with HA-eIF4G1 when eIF5 was overproduced (Fig. (Fig.4B),4B), suggesting that eIF5 binds to eIF4G not only in vitro (6) but also in vivo.
In order to examine the physiological relevance of the observed in vivo interactions between eIF4G and eIF1 or eIF5, we overproduced the latter in previously characterized temperature-sensitive (Ts−) eIF4G2 mutant strains and examined the growth of the resulting strains at restrictive or semipermissive temperatures. These mutant strains contain chromosomal deletions of both eIF4G-encoding loci, TIF4631 and TIF4632, and harbor a single-copy plasmid encoding mutant eIF4G2 as the sole source of eIF4G. The tif4632-1, tif4632-6, and tif4632-8 mutations shown in Fig. Fig.2B2B confer Ts− growth phenotypes that are suppressible by overproduced eIF4A (26). We found that the overexpression of eIF1 or eIF5 did not affect the Ts− growth of tif4632-1, tif4632-6, and tif4632-8 mutants at either restrictive (37°C) or semipermissive (35°C) temperatures (Fig. (Fig.5A,5A, lines 4 and 5 for tif4632-1; data not shown for tif4632-6 and -8). However, overexpression of eIF4A restored the growth of these mutants (Fig. (Fig.5A,5A, line 3 for tif4632-1; data not shown for tif4632-6 and -8), as reported previously (26).
Because the C-terminal half of eIF4G containing the HEAT domain interacts with eIF4A (21, 26) and also eIF1 and eIF5, we pondered whether overexpression of eIF1 or eIF5 would have a negative effect on the suppression of the tif4632 mutant phenotypes by overproduced eIF4A. To address this possibility, we constructed a plasmid overexpressing both eIF4A and eIF1 or eIF5, and we introduced it to the HEAT domain mutants mentioned above. We found that the suppression of tif4632-1 (Fig. (Fig.5A,5A, line 6), but not that of tif4632-6 or tif4632-8 (data not shown), by excess eIF4A was indeed reversed by cooverexpression of eIF1. We confirmed by immunoblot analyses that the level of eIF4A expression was not altered by cooverexpression of eIF1 (data not shown). To confirm that this effect is due to eIF1, we introduced a frameshift mutation, sui1Ω, into the eIF1 open reading frame present in the cooverexpression plasmid (Table (Table1).1). sui1Ω is unconditionally lethal (data not shown). As expected, sui1Ω in the cooverexpression plasmid restored the ability of intact eIF4A on the plasmid to suppress the tif4632-1 Ts− phenotype (Fig. (Fig.5A,5A, line 7). Likewise, the suppression of the tif4632-1 phenotype by eIF4A overexpression was reversed by cooverexpression of eIF5 (lines 8 and 9) without altering the level of eIF4A (data not shown). These results support the physiological relevance of eIF4G binding to overproduced eIF1 and eIF5 (Fig. (Fig.4),4), providing firm genetic evidence that eIF4G2 can interact with eIF1 and eIF5 in vivo, presumably at the HEAT domain.
The tif4632-430 mutation altering Leu-428 and Leu-429 to alanines is located outside of the HEAT domain and is deficient in interaction between eIF4G2 and eIF4E (36) (Fig. (Fig.2A).2A). During the course of the study, we observed that overexpression of eIF1 exacerbated the weak Ts− growth of the tif4632-430 mutant at the restrictive temperature (Fig. (Fig.5B,5B, line 14 versus 12). As eIF1 was overproduced from a high-copy plasmid carrying a 1.2-kb yeast chromosomal segment that contains other genes besides eIF1, we wished to verify that overexpression of eIF1 was responsible for this phenotype. For this purpose, we subcloned the eIF1 open reading frame under the GPD promoter carried on a high-copy-number vector and used the resulting plasmid for overexpression of eIF1. As expected, the effect of this plasmid on growth of the tif4632-430 mutant was identical to that of the eIF1 plasmid used in the previous experiment (Fig. (Fig.5B,5B, line 16). We conclude that the overproduction of eIF1 reduces the growth of the eIF4G2 mutant, which is deficient for eIF4G2-eIF4E association. We also found that overexpression of eIF1 did not affect the weak Ts− phenotypes of tif5-7A impairing eIF5 (4) or of tif34-1 impairing eIF3i (5), indicating that the synthetic effect of overexpression of eIF1 is specific for tif4632-430 (data not shown).
We then wished to determine whether this negative effect of eIF1 can be eliminated by cooverexpression of eIF4A or eIF5. Cooverexpression of both eIF4A (line 17) and eIF5 (line 18) reversed the synthetic phenotype between tif4632-430 and excess eIF1. The effect of eIF5 overexpression on this synthetic phenotype (line 18) is consistent with our finding that the binding of eIF4G2 to eIF1 and eIF5 is mutually exclusive (Fig. (Fig.3B3B to D); if association of the mutant eIF4G2 with eIF1 free of the ribosome is toxic, its prior association with eIF5 can prevent the toxic interaction due to the mutual exclusivity. Likewise, the counteractive effect of excess eIF4A on the synthetic phenotype (line 17) suggests that excess eIF4A outcompetes the binding of mutant eIF4G2 to eIF1 in vivo. Together with the effect of excess eIF1 on suppression of the tif4632-1 phenotype by excess eIF4A (Fig. (Fig.5A,5A, lines 6 and 7), the results suggest that the binding of eIF4A and at least eIF1 to the eIF4G HEAT domain is mutually exclusive. In conclusion, our results shown in Fig. Fig.5B5B provide a second piece of genetic evidence for in vivo interaction between eIF4G and eIF1.
The genetic data shown in Fig. Fig.55 suggest that the interactions of eIF4G with eIF1 and eIF4A, and those of eIF4G with eIF5 and eIF4A, are mutually competitive. In order to examine whether eIF1 directly competes with eIF4A for binding to the eIF4G HEAT domain, we allowed GST-eIF1 to bind 35S-eIF4G2439-914 in the presence of eIF4A in ~8-fold molar excess relative to GST-eIF1. Surface plasma resonance analyses indicate an equilibrium dissociation constant, Kd, of 3.4 ± 0.2 μM for eIF4A binding to eIF4G2439-914 (data not shown). Thus, the binding of eIF1, eIF5, and eIF4A were of similar strengths in all in vitro assays performed. Accordingly, the eightfold excess of eIF4A over GST-eIF1 when added to the reaction should reduce the GST-eIF1-35S-eIF4G2439-914 interaction ~10-fold at equilibrium if this interaction competes with the eIF4A-eIF4G2439-914 interaction. However, we found that this amount of eIF4A did not interfere with the interaction between GST-eIF1 and 35S-eIF4G2439-914 (Fig. (Fig.6,6, lane 5 versus lane 3), suggesting that the two interactions are not competitive. As eIF4A is an ATP-binding protein, we also examined the effect of ATP in the competition assay described above. However, the presence of ATP (1 mM) in the reaction did not affect the outcome of the experiments (Fig. (Fig.6,6, lane 7). Similarly, we found that the binding of GST-eIF5241-405 to eIF4G does not compete with the eIF4G-eIF4A interaction in the presence or absence of ATP (Fig. (Fig.6,6, lanes 6 and 8). We also found that the binding of GST-eIF4G2439-914 (~1 μg) to 35S-eIF1 was not inhibited by excess eIF4A (50 μg), even though eIF4A was present in large excess over GST-eIF4G2439-914 and can bind to the latter (data not shown). Therefore, the purified eIF4A did not compete with eIF1, or eIF5, for the interaction with the eIF4G2 HEAT domain under the conditions we examined. Because we confirmed that tif4632-1 completely abolishes GST-eIF4G2439-914-eIF4A interaction in vitro (data not shown; see also reference 26), we were not able to address the counteractive effect of eIF1 or eIF5 on this already weakened interaction. These results indicate that direct physical competition for binding to the HEAT domain may not underlie the functional competition between eIF1 or eIF5 and eIF4A observed in vivo. The idea of competition between eIF1 or eIF5 and eIF4A for eIF4G2 must be addressed in experiments including eIF4B, eIF3, or the entire eIF4G, either wild type or mutant (see Discussion).
So far, we showed that eIF4G interacts with eIF1 and eIF5 in vitro (Fig. (Fig.11 to to3)3) and in vivo (Fig. (Fig.4),4), and we confirmed the biological significance of the in vivo interactions of eIF4G with overproduced eIF1 or eIF5 by using genetic approaches (Fig. (Fig.5).5). Next we turned our attention to examining the direct role of the eIF4G HEAT domain in the 48S complex during stringent AUG selection. For this purpose, we tested if the known eIF4G2 mutations altering residues in the HEAT domain (Fig. (Fig.2B)2B) allow increased translation from a non-AUG start codon. We transformed the tif4632-1, tif4632-6, and tif4632-8 mutants with a UUG-his4::lacZ plasmid and assayed β-galactosidase activity produced in each transformant. The UUG-his4::lacZ construct contains two UUG codons in frame with lacZ, one at codon 1 and the other at codon 3 of his4::lacZ. It was shown that among non-AUG codons, UUG is used preferentially in different Sui− mutants (16). As shown in Fig. Fig.7A,7A, the tif4632-1 (column 2) and tif4632-6 (column 3) mutants had 1.8- and 2.2-fold higher expression of UUG-his4::lacZ, respectively, than the isogenic wild-type strain. As a positive control, we found that a known Sui− mutation in eIF5 (E. Hannig and C. Curtis, personal communications) led to 4.9-fold higher UUG-his4::lacZ expression than its isogenic wild type (Fig. (Fig.7A,7A, columns 5 and 6).
In order to test whether the UUG-his4::lacZ expression observed with the tif4632 mutants depends upon the UUG start codon, we assayed his4::lacZ expression from a second reporter containing AUU and UUA at codons 1 and 3, respectively. As shown in Fig. Fig.7B,7B, lacZ expression from this reporter was ~60% of the values obtained with UUG-his4::lacZ in the wild-type strains (Fig. (Fig.7A7A and B, columns 1 and 5) and was not altered at all by the tif4632 mutations or the control SUI5 mutation. (Note the same scale used for Fig. Fig.7A7A and B.) Therefore, the UUG codon appears to be preferentially used as a start codon in tif4632-1 and tif4632-6 mutants, as observed previously for a SUI5 mutant (16).
To determine whether the increased UUG-his4 expression is accompanied by an increase in translation from the canonical AUG codon, the AUG-his4::lacZ plasmid containing AUG at codon 1 was also analyzed in parallel. As we used a synthetic complete (SC) medium to grow yeast prior to β-galactosidase assays, his4::lacZ transcription should not be derepressed by Gcn4p via the general amino acid control pathway. Consistently, we found that expression of AUG-his4::lacZ was relatively low and was not altered by tif4632-1 or tif4632-8 (Fig. (Fig.7C,7C, columns 1, 2, and 4). However, AUG-his4::lacZ translation was increased twofold with tif4632-6 (Fig. (Fig.7C,7C, column 3), suggesting that this mutation increases the frequency of initiation from both UUG and AUG codons. After normalizing for AUG-his4::lacZ expression, the UUG suppression activity (the percentage of the value in Fig. Fig.7A7A relative to that in C) still increased 1.9-fold in tif4632-1 compared to the isogenic wild type (Fig. (Fig.7D).7D). (The mutant Sui− phenotype is more obvious in the experiment shown below in Table Table3,3, as the UUG/AUG expression ratio for the vector control tif4632-1 transformant was 3.2-fold higher than that for the wild type.) The control SUI5 mutation increased the UUG suppression activity 5.2-fold (Fig. (Fig.7D).7D). We also found that the tif4632-430 mutation, defective in eIF4E binding to eIF4G2, did not change LacZ expression from either the AUG-his4 or UUG-his4 reporter compared to wild type (data not shown). Thus, the tif4632-1 mutation specifically increases translation from a UUG codon and hence shows a mild Sui− phenotype. Since we grew tif4632 mutants at a permissive temperature prior to the assays, the observed phenotype should not be due to a secondary effect of limiting mRNA binding.
We also tested if the mild Sui− phenotype observed with tif4632-1 is suppressible by overexpression of eIF1 and eIF5. Table Table33 summarizes the results of a β-galactosidase assay performed with tif4632-1 transformants carrying the his4::lacZ fusion plasmid and compatible eIF1 or eIF5 overexpression plasmid, together with appropriate control transformants. We found that overexpression of eIF1 or eIF5 did not alter the low UUG/AUG expression ratio (3.5 to 3.8%) of the wild-type strain (lines 1 to 6, column 6). Interestingly, overexpression of eIF1, but not that of eIF5, significantly reduced the UUG/AUG expression ratio that was increased threefold by tif4632-1 (lines 7 to 12, column 6). These results strongly suggest that a reduced interaction with eIF1, as observed in Fig. Fig.2B,2B, at least partially accounts for the Sui− phenotype caused by tif4632-1.
Finally, we examined whether the reduced interaction between eIF4G and eIF1 could account for the phenotypes of previously known Sui− mutants mapping in eIF1 (41). For this purpose, we first conducted GST pull-down assays with an eIF1 mutant carrying sui1-1 (D83G). To our surprise, sui1-1 reduced the binding of 35S-eIF1 to GST-eIF4G2439-914 by fourfold, but not its binding to GST-eIF3c1-156 (Fig. (Fig.8A,8A, top and second panel). As a control, GST-eIF4G2439-914, but not GST-eIF3c1-156, bound specifically to 35S-eIF4A (third panel), confirming the previously identified interaction (26). These results indicate that a known Sui− mutation in eIF1 reduces its binding to eIF4G2 in vitro.
Next, we tested the effect of sui1-1 on the phenotype we observed with tif4632-430. We found that overexpression of the wild-type SUI1 allele, but not of the sui1-1 allele, exacerbated the weak Ts− phenotype of tif4632-430 (Fig. (Fig.8B,8B, lines 3 and 4). Immunoblotting with anti-eIF1 antibodies indicated that both SUI1 and sui1-1 alleles overproduce equivalent levels of eIF1 (Fig. (Fig.8C).8C). Thus, the lack of synthetic phenotype with sui1-1 is due to its functional defect, presumably in the interaction with eIF4G. Based on these results, we conclude that the interaction of the eIF4G HEAT domain with at least eIF1 is important for the initiation complex to locate the first AUG codon accurately.
The HEAT domain of mammalian eIF4GII is comprised of five stacked pairs of antiparallel α-helices, with each pair corresponding to a HEAT repeat, and it interacts with eIF4A (2, 21). In this report, we showed that the HEAT domain and flanking residues of yeast eIF4G2 are required for its binding to the AUG recognition factors eIF1 and eIF5 (Fig. (Fig.11 and and2).2). We also provided ample genetic evidence that eIF1 functionally interacts with the eIF4G HEAT domain (Fig. (Fig.5).5). Furthermore, we found that a HEAT domain mutation in eIF4G2 (tif4632-1) led to a mild increase in translation from a UUG codon (Fig. (Fig.7),7), partially suppressible by eIF1 overexpression (Table (Table3),3), and that the sui1-1 mutation reduced the binding of eIF1 to eIF4G2, both by in vitro binding assays and by genetic experiments (Fig. (Fig.8).8). These results together support a role for the eIF4G HEAT domain in the AUG recognition step in which eIF1 and eIF5 participate.
Our deletion and mutational analyses of yeast eIF4G2 interaction in vitro (Fig. (Fig.2)2) are consistent with the idea that its C-terminal half is composed of a continuous structure (HEAT domain) required for its optimal binding to different binding partners, eIF5 and eIF1 in this case. The requirement for the 118-amino-acid-long N-terminal extension is consistent with the strong conservation of the segment in eukaryotes (25), as summarized in Fig. Fig.2A,2A, and the isolation of a Ts− mutation designated tif4632-8 that alters three amino acids in this domain and a fourth in the HEAT domain (6) (also Fig. Fig.2B).2B). On the other hand, the C-terminal two-thirds of the carboxyl half of eIF4G2 (up to residue 578) were largely dispensable for the interaction with eIF5, but deletion of residues 846 to 914 just C terminal to the HEAT domain reduced substantially the interaction with eIF1 (Fig. (Fig.2A).2A). Thus, the first HEAT repeat and the N-terminal extension found in eIF4G2439-654 may form a minimal structure sufficient for eIF5 binding, but the eIF1 binding requires an extended HEAT repeat structure. As a HEAT domain is composed of 3 to 22 HEAT repeats (or 6 to 44 α-helices) (6), the N- and C-terminal extensions required for the tightest binding of eIF1 and eIF5 may simply contain additional HEAT repeats.
Many pieces of genetic evidence have linked eIF1 to the heart of the decoding site of the 40S ribosome. Accordingly, it was proposed that eIF1 is a general monitor of codon-anticodon pairing fidelity, being bound near the ribosomal P-site (9, 41). Our finding of the mild Sui− phenotype for tif4632-1 (Fig. (Fig.7),7), its partial suppression by eIF1 overexpression (Table (Table3),3), and a reduced eIF1-eIF4G2 interaction by sui1-1 (Fig. (Fig.8)8) support the model that the multiple, simultaneous interactions of eIF1 with eIF4G and other factors are important to position eIF1 securely to its functional site on the ribosome. It is possible that loose positioning of eIF1 caused by sui1-1 or tif4632-1 mutation promotes earlier release of initiation factors even at noncognate AUG pairing, thereby relaxing stringent start codon selection.
Because Fig. Fig.3B3B to D and and5B5B (line 18) suggest that the observed interactions of eIF4G with eIF1 and eIF5 are mutually exclusive, we hypothesize that eIF4G interacts with eIF1 and eIF5-CTD at different steps. eIF4G may bind first to eIF5-CTD, because evidence suggests that eIF5-CTD stimulates mRNA recruitment to the ribosome (6). Subsequently, eIF4G may bind to eIF1 during the process of AUG recognition for precise juxtaposition of the latter in the initiation complex, as illustrated in Fig. Fig.9.9. We also reported previously that the eIF5-eIF4G interaction is exclusive with the eIF2β-eIF5 interaction that mediates formation of the MFC prior to its recruitment to the ribosome (6). Thus, the initiation factors assembled on the ribosome may undergo an isomerization (without altering their relative positions) as they proceed through different steps in the pathway from 43S to 48S complexes.
The relatively low affinity deduced for eIF4G interactions with eIF1 and eIF5 (Kd, ~10 μM) (Fig. (Fig.1)1) suggests that these interactions mainly occur in the 48S complex at the concentrations of these factors found in the cell. Because the steady-state level of 48S complexes is much lower than that of 43S complexes containing only the MFC components (3), it was difficult to observe these interactions in vivo by coimmunoprecipitation without overproducing eIF1 or eIF5 (Fig. (Fig.4).4). The physiological relevance of these interactions, however, was confirmed by finding that eIF1 or eIF5 overexpression reversed the suppression of the Ts− phenotype of tif4632-1 (altering the HEAT domain) by excess eIF4A (Fig. (Fig.5A)5A) and that eIF1 overexpression exacerbated the weak Ts− phenotype of tif4632-430, which was reversed by cooverexpression of eIF4A and eIF5 (Fig. (Fig.5B5B).
The eIF5B (domain IV)-eIF1A (CTD) interaction provides the precedence for a well-characterized, low-affinity interaction in translation initiation. Like the effect of eIF1 on tif4632-430 (Fig. (Fig.5B),5B), overexpression of eIF1A exacerbates the slow-growth phenotype conferred by eIF5B-domain IV deletion in yeast (7). It is proposed that this interaction is critical for coupling the release of eIF1A to that of eIF5B upon the hydrolysis of GTP bound to the latter; thus, tighter association of eIF1A with the 40S ribosome by mass action would severely impede 60S subunit joining already impaired by the lack of eIF5B domain IV (27). The solution structure of the eIF5B/eIF1A complex has just been solved at an atomic resolution, and computer-aided docking studies suggest an interaction between the 40S ribosome and the eIF5B/eIF1A complex that is consistent with this model (22).
It is intriguing that eIF1 has been shown to bind to the Upf2 subunit of the surveillance complex at the “eIF4G homology domain” (23), which in fact is predicted to fold into a HEAT domain based on the mammalian eIF4GII structure (21). The surveillance complex is proposed to bind to the ribosome at translation termination and scan the mRNA downstream towards the poly(A) sequence for a specific cis element, thereby programming the mRNA for nonsense-mediated decay when the complex binds to the cis element (40). A role of the surveillance complex is also proposed for monitoring translation fidelity during elongation (also reviewed in reference 9). As Upf2 binds an RNA helicase (Upf1), a HEAT domain complexed with a helicase and eIF1 may be a more universal vehicle of regulating ribosome function during all three steps of translation.
Although the genetic data in Fig. Fig.55 suggest direct competition between eIF1 and eIF4A, or between eIF5 and eIF4A, for binding to the eIF4G HEAT domain, we were not able to obtain data in support of this model, as in vitro binding studies suggested rather simultaneous binding of the eIF4G HEAT domain fragment with eIF4A and eIF1 or eIF5 (Fig. (Fig.6).6). However, mammalian eIF4B clearly stimulates formation of the 48S complex positioned at the start codon (28) and is conserved in yeast (1). In addition, mammalian eIF4G binds cooperatively to eIF3 and eIF4A (18). Thus, it is possible that the conformation of the HEAT domain may alter upon the association of eIF4G with eIF3, eIF4B, and/or mRNA, so that the binding surfaces for eIF1/eIF5 and eIF4A become mutually exclusive. Such a conformational change may ensure the correct order of factor assembly of eIF4G-containing complexes, with eIF4A in the cap-binding and unwinding steps preceding the function of eIF1 and eIF5 in AUG recognition.
Although this model is attractive, an alternative explanation for the phenotypic competition is that it occurs only when the function of eIF4G2 is compromised by specific mutations, tif4632-1 and tif4632-430 (Fig. (Fig.2A).2A). Indeed, overexpression of eIF1 (Fig. (Fig.5B,5B, line 11) or eIF5 (data not shown) does not inhibit the growth of wild-type yeast (encoding wild-type eIF4Gs), nor did it have any effect on suppression of two other Ts− alleles, tif4632-6 and tif4632-8, by eIF4A (data not shown). Thus, the phenotypic competition in Fig. Fig.55 might be explained if we assume that tif4632-1 is additionally defective in binding to factor X (e.g., eIF3 or eIF4B) that stimulates the eIF4A-eIF4G interaction, such that the ability of eIF4G to bind eIF4A is severely impaired. In this event, even a subtle conformational change induced by binding of the mutant eIF4G2 to overproduced eIF1 or eIF5 would abolish the binding of eIF4A to an extent that cannot be remedied by eIF4A overexpression (Fig. (Fig.5A,5A, rows 6 and 9). In the case of tif4632-430, we suggest that the binding of eIF1 and factor X to the HEAT domain of the mutant eIF4G2 is mutually exclusive. Accordingly, the mutant eIF4G2 bound to overproduced eIF1 would not bind factor X, thereby reducing the binding to eIF4A, and produce a synthetic growth defect that is suppressible by eIF4A overexpression (Fig. (Fig.5B,5B, rows 14, 16, and 18).
Deciphering the relationship between binding of eIF4A and eIF1 to eIF4G is particularly important, now that the roles of eIF1 and eIF4A in scanning have been defined separately in AUG recognition and mRNA unwinding, respectively, by an elegant biochemical experiment. eIF4 factors and ATP are dispensable for the 43S complex containing eIFs 1, 2, 3, and 1A to locate the first AUG in an mRNA with an unstructured leader, but they are required for it to locate the first AUG in an mRNA with even a weak stem-loop in the leader region (29).
We are greatly indebted to Alan Sachs for generous gifts of materials and permission to use YAS1998, YAS1999, and YAS2000 prior to publication and to Ernie Hannig and Cynthia Curtis for a SUI5 LEU2 plasmid prior to publication and technical advice on β-galactosidase assays. We also thank Tom Donahue and Stu Peltz for timely gifts of plasmids, Ashik Srinivasan for technical help, and Beth Montelone and other members of the KSU MCDB Program for advice and discussion.
This work was supported by NIH COBRE award 1 P20 RR15563, matching support from the State of Kansas and KSU, NIH grant GM64781 to K.A., and Wellcome Trust funding to J.E.G.M.
Hui He, Tobias von der Haar, C. Ranjit Singh, and Miki Ii contributed equally to this work.